GPUnet: Networking abstractions for GPU programs

Mark Silberstein; Sangman Kim; Seonggu Huh; Xinya Zhang; Yige Hu; Amir Wated; Emmett Witchel

doi:https://doi.org/10.1145/2963098

GPUnet: Networking abstractions for GPU programs

Mark Silberstein, Sangman Kim, Seonggu Huh, Xinya Zhang, Yige Hu, Amir Wated, Emmett Witchel

Electrical and Computer Engineering

Technion - Israel Institute of Technology

Research output: Contribution to journal › Article › peer-review

Abstract

Despite the popularity of GPUs in high-performance and scientific computing, and despite increasingly general-purpose hardware capabilities, the use of GPUs in network servers or distributed systems poses significant challenges. GPUnet is a native GPU networking layer that provides a socket abstraction and high-level networking APIs for GPU programs. We use GPUnet to streamline the development of high-performance, distributed applications like in-GPU-memory MapReduce and a new class of low-latency, high-throughput GPU-native network services such as a face verification server.

Original language	English
Article number	9
Journal	ACM Transactions on Computer Systems
Volume	34
Issue number	3
DOIs	https://doi.org/10.1145/2963098
State	Published - Sep 2016

Keywords

Accelerators
GPGPUs
Network servers
Operating systems design

All Science Journal Classification (ASJC) codes

General Computer Science

Access to Document

https://doi.org/10.1145/2963098

Cite this

@article{d5391d0bcb864fc39ae8f4ca2be7e38c,

title = "GPUnet: Networking abstractions for GPU programs",

abstract = "Despite the popularity of GPUs in high-performance and scientific computing, and despite increasingly general-purpose hardware capabilities, the use of GPUs in network servers or distributed systems poses significant challenges. GPUnet is a native GPU networking layer that provides a socket abstraction and high-level networking APIs for GPU programs. We use GPUnet to streamline the development of high-performance, distributed applications like in-GPU-memory MapReduce and a new class of low-latency, high-throughput GPU-native network services such as a face verification server.",

keywords = "Accelerators, GPGPUs, Network servers, Operating systems design",

author = "Mark Silberstein and Sangman Kim and Seonggu Huh and Xinya Zhang and Yige Hu and Amir Wated and Emmett Witchel",

note = "Publisher Copyright: {\textcopyright} 2016 ACM.",

year = "2016",

month = sep,

doi = "https://doi.org/10.1145/2963098",

language = "الإنجليزيّة",

volume = "34",

journal = "ACM Transactions on Computer Systems",

issn = "0734-2071",

publisher = "Association for Computing Machinery (ACM)",

number = "3",

}

TY - JOUR

T1 - GPUnet

T2 - Networking abstractions for GPU programs

AU - Silberstein, Mark

AU - Kim, Sangman

AU - Huh, Seonggu

AU - Zhang, Xinya

AU - Hu, Yige

AU - Wated, Amir

AU - Witchel, Emmett

PY - 2016/9

Y1 - 2016/9

N2 - Despite the popularity of GPUs in high-performance and scientific computing, and despite increasingly general-purpose hardware capabilities, the use of GPUs in network servers or distributed systems poses significant challenges. GPUnet is a native GPU networking layer that provides a socket abstraction and high-level networking APIs for GPU programs. We use GPUnet to streamline the development of high-performance, distributed applications like in-GPU-memory MapReduce and a new class of low-latency, high-throughput GPU-native network services such as a face verification server.

AB - Despite the popularity of GPUs in high-performance and scientific computing, and despite increasingly general-purpose hardware capabilities, the use of GPUs in network servers or distributed systems poses significant challenges. GPUnet is a native GPU networking layer that provides a socket abstraction and high-level networking APIs for GPU programs. We use GPUnet to streamline the development of high-performance, distributed applications like in-GPU-memory MapReduce and a new class of low-latency, high-throughput GPU-native network services such as a face verification server.

KW - Accelerators

KW - GPGPUs

KW - Network servers

KW - Operating systems design

UR - http://www.scopus.com/inward/record.url?scp=84989353495&partnerID=8YFLogxK

U2 - https://doi.org/10.1145/2963098

DO - https://doi.org/10.1145/2963098

M3 - مقالة

SN - 0734-2071

VL - 34

JO - ACM Transactions on Computer Systems

JF - ACM Transactions on Computer Systems

IS - 3

M1 - 9

ER -

GPUnet: Networking abstractions for GPU programs

Abstract

Keywords

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this